中文版 | English
CSpace
Multi-Window Local Region Contrast Adjustment and Region Growing for Retinal Vessel Segmentation
Wang, Weihua1,3,4; Wang, Weiqing2; Wu, Wenyuan3; Zhang, Jingzhong3
2018-10-01
摘要Retinal vessel images provide a wealth of information for the diagnosis of several medical pathologies. Accurate segmentation of blood vessel image from fundus image is crucial for pathological analysis. To address the issues of the fine vessel branches that are overshadowed by the major vessels, high-brightness areas with low contrast in a local region, a novel multi-window local region contrast adjustment and region growing for retinal vessel enhancement and segmentation is proposed in this paper. Firstly, the Gabor wavelet is used for enhancing blood vessels. Then, a local region contrast adjustment method with a window scale is proposed to further enhance the fine vessels with the low contrast between the vessel and background in a local region. Finally, an iterative region growing method based on adaptive multi-threshold is presented to extract the blood vessels from an enhanced fundus image, the iterative process can be stopped automatically. The performance of this algorithm is evaluated on two publicly fundus image databases DRIVE and STARE by five different metrics. The proposed method reaches an average segmentation accuracy, specificity and sensitivity of 0.9443, 0.9743 and 0.7516 for the DRIVE database respectively, and 0.9464, 0.9700 and 0.7566 for the STARE databases respectively. The experiment results demonstrated that our proposed method with an automated termination condition improved the true segmentation of the fine vessels and decreases the false segmentation of the vessels in low contrast regions compared to the existing unsupervised methods.
关键词Local Contrast Retinal Vessel Segmentation Medical Image Processing Region Growing
DOI10.1166/jmihi.2018.2503
发表期刊JOURNAL OF MEDICAL IMAGING AND HEALTH INFORMATICS
ISSN2156-7018
卷号8期号:8页码:1554-1565
通讯作者Wang, Weiqing()
收录类别SCI
WOS记录号WOS:000449743300003
语种英语